Smart Edge Devices Research Articles

A decentralised federated learning scheme for heterogeneous devices in cognitive IoT

Cognitive Internet of Things (IoT) technologies typically rely on substantial data collected from edge devices for data analysis and decision-making. However this reliance often leads to the inadvertent exposure of private data from smart edge devices. Federated learning (FL) is a distributed machine learning framework that protects user privacy by performing collaborative training without uploading private data. Nevertheless, applying classical FL to cognitive IoT systems to preserve privacy preservation faces significant challenges, such as central server failure and communication burden. Furthermore, when edge devices with heterogeneous data and systems participate in federated training, the learning process becomes slow and the performance of edge devices is compromised. To address these challenges, we propose a decentralised FL framework for cognitive IoT, termed DFL–MKF. In DFL– MKF, the centralised server is eliminated and each edge device is dynamically connected. We initialised models for edge devices based on computational and storage capabilities to accommodate system heterogeneity. Edge devices learned from the knowledge of multiple neighbours via knowledge transfer, and knowledge fusion was employed to aggregate the knowledge of multiple neighbours, thereby improving the performance of local models, and addressing data heterogeneity. Comprehensive experiments were performed on three image classification tasks. The results of these experiments demonstrate that the proposed method achieved superior performance compared to various baselines and improved communication efficiency.

A 184-μW Error-Tolerant Real-Time Hand Gesture Recognition System With Hybrid Tiny Classifiers Utilizing Edge CNN

This article proposes a low-power real-time hand gesture recognition (HGR) system with high recognition accuracy for smart edge devices. This design balances accuracy and power consumption by utilizing computation-efficient hybrid classifiers assisted with a majority voting scheme. By combining the recognition results of consecutive frames, the HGR system shows improved immunity to misclassification. In addition, the compressed input data before high-level processing dramatically reduce the on- chip memory and computational load. The proposed Edge-convolutional neural network (CNN) core with interactable processing engines reduces the memory accessing and the feature register toggling rate by 27% and 50%, respectively. The sequence analyzer based on majority voting improves the static and dynamic gesture recognition accuracy by ~7% and ~8% only with 9.4% hardware overhead. The test chip was fabricated in 65-nm CMOS technology, occupying the area of 1 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times $ </tex-math></inline-formula> 1.5 mm2. It consumes the lowest power of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$184~\mu \text{W}$ </tex-math></inline-formula> at 25 MHz and 0.6 V. The proposed HGR system can recognize six static gestures and 24 dynamic hand gestures with an average accuracy of 87.25%–95% and 85.4%–94.9%, respectively.

Decentralized Deep Learning for Multi-Access Edge Computing: A Survey on Communication Efficiency and Trustworthiness

Wider coverage and a better solution to a latency reduction in 5G necessitate its combination with multi-access edge computing technology. Decentralized deep learning (DDL), such as federated learning and swarm learning, as a promising solution to privacy-preserving data processing for millions of smart edge devices leverages distributed computing of multilayer neural networks within the networking of local clients, without disclosing the original local training data. Notably, in industries such as finance and healthcare, where sensitive data of transactions and personal medical records are cautiously maintained, DDL can facilitate the collaboration among these institutes to improve the performance of trained models while protecting the data privacy of participating clients. In this survey article, we demonstrate the technical fundamentals of DDL that benefit many walks of society through decentralized learning. Furthermore, we offer a comprehensive overview of the current state of the art in the field by outlining the challenges of DDL and the most relevant solutions from novel perspectives of communication efficiency and trustworthiness.

Hardware Solutions for Low-Power Smart Edge Computing

The edge computing paradigm for Internet-of-Things brings computing closer to data sources, such as environmental sensors and cameras, using connected smart devices. Over the last few years, research in this area has been both interesting and timely. Typical services like analysis, decision, and control, can be realized by edge computing nodes executing full-fledged algorithms. Traditionally, low-power smart edge devices have been realized using resource-constrained systems executing machine learning (ML) algorithms for identifying objects or features, making decisions, etc. Initially, this paper discusses recent advances in embedded systems that are devoted to energy-efficient ML algorithm execution. A survey of the mainstream embedded computing devices for low-power IoT and edge computing is then presented. Finally, CYSmart is introduced as an innovative smart edge computing system. Two operational use cases are presented to illustrate its power efficiency.

Self-Healing of Semantically Interoperable Smart and Prescriptive Edge Devices in IoT

Smart homes enhance energy efficiency without compromising residents’ comfort. To support smart home deployment and services, an IoT network must be established, while energy-management techniques must be applied to ensure energy efficiency. IoT networks must perpetually operate to ensure constant energy and indoor environmental monitoring. In this paper, an advanced sensor-agnostic plug-n-play prescriptive edge-to-edge IoT network management with micro-services is proposed, supporting also the semantic interoperability of multiple smart edge devices operating in the smart home network. Furthermore, IoT health-monitoring algorithms are applied to inspect network anomalies taking proper healing actions/prescriptions without the need to visit the residency. An autoencoder long short-term memory (AE-LSTM) is selected for detecting problematic situations, improving error prediction to 99.4%. Finally, indicative evaluation results reveal the mitigation of the IoT system breakdowns.

Industrial Edge Computing: Architecture, Challenges, Applications

Edge computing is the deployment of network operations and data-handling activities toward more solitary sources of data collecting or storage. Industrial edge computing, which includes industrial applications, describes the method of controlling data-handling operations using distinct data sources, such as smart edge devices. As a result, whether collecting, processing, or evaluating data in smart factories, IIoT devices and smart equipment do not need to contact centralised cloud platforms. Information can also be sent to the cloud for additional analysis or integration into a larger system, depending on how important the data analysis performed at the edge is. Industrial edge computing is a term used to describe a distributed platform that combines communication, processing, and storage capabilities to run real-time applications that may be accessed directly from the cloud. The Industrial Internet of Things (IIoT) revolution has raised demands for flexible connectivity, realtime control, data optimization, intelligent applications, security, and privacy protection. Edge computing nodes serve as intelligent gateways for assets, services, and systems, bridging the gap between the real and digital worlds.

PCFed: Privacy-Enhanced and Communication-Efficient Federated Learning for Industrial IoTs

Federated learning (FL) is capable of analyzing tremendous data from smart edge devices in Industrial Internet of Things (IIoTs), empowering numerous industrial applications. However, the increasing privacy concerns and deployment costs of IIoT environment have been posing new challenges for FL. This article proposes PCFed, a novel privacy-enhanced and communication-efficient FL framework to provide higher model accuracy with rigorous privacy guarantees and great communication efficiency. In particular, we develop a sampling-based intermittent communication strategy via a PID (proportional, integral, and derivative) controller on the cloud server to adaptively reduce the communication frequency. In addition, we design a budget allocation mechanism to balance the tradeoff between model accuracy and privacy loss. Then, we develop PCFed+, an enhanced variant for PCFed, with further consideration of infinite data streams on edge servers. Extensive experiments demonstrate that both PCFed and PCFed+ can significantly outperform existing schemes, in terms of communication efficiency, privacy protection, and model accuracy.

Convolutional Echo‐State Network with Random Memristors for Spatiotemporal Signal Classification

The unprecedented development of Internet of Things results in the explosion of spatiotemporal signals generated by smart edge devices, leading to a surge of interest in real‐time learning of such data. This imposes a big challenge to conventional digital hardware because of physically separated memory and processing units and the transistor scaling limit. Memristors are deemed a solution for efficient and portable deep learning. However, their ionic resistive switching incurs large programming stochasticity and energy, compromising their advantages in real‐time learning spatiotemporal signals. To address the aforementioned issues, we propose a novel hardware–software codesign. Hardware‐wise, the stochasticity in memristor programming is leveraged to produce random matrices for efficient in‐memory computing. Software‐wise, random convolutional‐pooling architectures are integrated with echo‐state networks that compute with the hardware random matrices and make real‐time learning affordable. The synergy of the hardware and software not only improves the performance over conventional echo‐state networks, that is, 90.94% and 91.67% (compared to baselines 88.33% and 62.50%), but also retains 187.79× and 93.66× improvement of energy efficiency compared to the digital alternatives on the representative Human Activity Recognition Using Smartphones (HAR) and CRICKET datasets, respectively. These advantages make random convolutional echo‐state network (RCESN) a promising solution for the future smart edge hardware.

SmartDefense: A distributed deep defense against DDoS attacks with edge computing

The growing number of IoT edge devices have inflicted a change in the cyber-attack space. The DDoS attacks, in particular, have significantly increased in magnitude and intensity. Of the existing DDoS solutions, while the destination-based defense mechanisms incur high false positives due to the seemingly legitimate nature of the attack traffic, defense mechanisms implemented at the source alone do not suffice due to the lack of visibility into ongoing DDoS attacks. This paper proposes a distributed DDoS detection and mitigation framework, SmartDefense, based on edge computing approaches towards detecting and mitigating DDoS attacks at and near the source. By mitigating the DDoS attacks near the source, SmartDefense significantly reduces unnecessary bandwidth otherwise consumed by DDoS traffic going from residential edge networks to the ISP edge network.Furthermore, SmartDefense demonstrates how ISPs can detect botnet devices in their customer’s network by having smart edge devices pass attributes that are processed by the botnet detection engine at the provider’s edge. The evaluation of this work shows that SmartDefense can improve the detection and mitigation rate, with over 90% of DDoS traffic caught at the source and over 97.5% of remaining DDoS traffic caught at the provider’s edge. Our experiments also demonstrate how using a botnet detection engine can further reduce the DDoS traffic by up to 51.95% by facilitating ISPs to detect bot devices in their customers’ edge network.

Developing smart devices with automated Machine learning Approach: A review

In the current era, Artificial Intelligence is playing a major role in smart devices. Apart from incorporating intelligence in the devices, it has extended its range beyond that. Now new AI development tools such as automated machine learning (AutoML) approach are increasing rapidly because of its performance on smart devices. AutoML is used to design a machine learning model with self-adapting capability which can perform the task in less time with advanced skills. This approach does not need any explicit coding and the data science experts but this can build smart sensor codes that will get executed automatically on the microcontrollers embedded within the IoT devices. Due to these capabilities, the developers can build efficient smart devices with real-time responsiveness, security, and data privacy without the help of machine learning experts. In this paper, we will show how AutoML will be used to create smart de-vices. The objective of this paper is to develop the skills that will help to develop smart edge devices which include planning, design-ing, implementation, and the evaluation of the smart IoT applications.

Secure Data Sharing with Efficient Key Update for Industrial Cloud-based Access Control

Sharing data in industrial cloud requires both secure and flexibility management of data shared among users in the industry. Users should have ability to access authorized portion of data from smart edge devices with integrated secure protocol for interacting with cloud storage. Among cryptographic-based access control solutions, ciphertext-policy attribute-based encryption (CP-ABE) is recognized as a suitable solution for supporting secure and fine-grained data access control for outsourced data. However, the attribute revocation is a major drawback of CP-ABE since it introduces subsequent costs such as ciphertext re-encryption, user key re-generation, and key re-distribution. Existing revocable CP-ABE based access control models have focused on minimizing the cost of communication and computation cost for ciphertext re-encryption. Unfortunately, the key update issue is subject to re-key generation as traditional CP-ABE does. In essence, key update becomes crucial when there are a high number of users accessing shared data in the cloud. In this paper, we propose a lightweight access control model featured with the efficient key updating scheme. Finally, the performance evaluation is conducted to demonstrate the efficiency of our proposed scheme.

EffCNet: An Efficient CondenseNet for Image Classification on NXP BlueBox

Intelligent edge devices with built-in processors vary widely in terms of capability and physical form to perform advanced Computer Vision (CV) tasks such as image classification and object detection, for example. With constant advances in the field of autonomous cars and UAVs, embedded systems and mobile devices, there has been an ever-growing demand for extremely efficient Artificial Neural Networks (ANN) for real-time inference on these smart edge devices with constrained computational resources. With unreliable network connections in remote regions and an added complexity of data transmission, it is of an utmost importance to capture and process data locally instead of sending the data to cloud servers for remote processing. Edge devices on the other hand, offer limited processing power due to their inexpensive hardware, and limited cooling and computational resources. In this paper, we propose a novel deep convolutional neural network architecture called EffCNet which is an improved and an efficient version of CondenseNet Convolutional Neural Network (CNN) for edge devices utilizing self-querying data augmentation and depthwise separable convolutional strategies to improve real-time inference performance as well as reduce the final trained model size, trainable parameters, and Floating-Point Operations (FLOPs) of EffCNet CNN. Furthermore, extensive supervised image classification analyses are conducted on two benchmarking datasets: CIFAR-10 and CIFAR-100, to verify real-time inference performance of our proposed CNN. Finally, we deploy these trained weights on NXP BlueBox which is an intelligent edge development platform designed for self-driving vehicles and UAVs, and conclusions will be extrapolated accordingly.

O3BNN-R: An Out-of-Order Architecture for High-Performance and Regularized BNN Inference

Binarized Neural Networks (BNN), which significantly reduce computational complexity and memory demand, have shown potential in cost- and power-restricted domains, such as IoT and smart edge-devices, where reaching certain accuracy bars is sufficient and real-time is highly desired. In this article, we demonstrate that the highly-condensed BNN model can be shrunk significantly by dynamically pruning irregular redundant edges. Based on two new observations on BNN-specific properties, an out-of-order (OoO) architecture, O3BNN-R, which can curtail edge evaluation in cases where the binary output of a neuron can be determined early at runtime during inference, is proposed. Similar to instruction level parallelism (ILP), fine-grained, irregular, and runtime pruning opportunities are traditionally presumed to be difficult to exploit. To further enhance the pruning opportunities, we conduct an algorithm/architecture co-design approach where we augment the loss function during the training stage with specialized regularization terms favoring edge pruning. We evaluate our design on an embedded FPGA using networks that include VGG-16, AlexNet for ImageNet, and a VGG-like network for Cifar-10. Results show that O3BNN-R without regularization can prune, on average, 30 percent of the operations, without any accuracy loss, bringing 2.2× inference-speedup, and on average 34× energy-efficiency improvement over state-of-the-art BNN implementations on FPGA/GPU/CPU. With regularization at training, the performance is further improved, on average, by 15 percent.

Benchmark of the Compute-in-Memory-Based DNN Accelerator With Area Constraint

Compute-in-memory (CIM) is a promising computing paradigm to accelerate the inference of deep neural network (DNN) algorithms due to its high processing parallelism and energy efficiency. Prior CIM-based DNN accelerators mostly consider full custom design, which assumes that all the weights are stored on-chip. For lightweight smart edge devices, this assumption may not hold. In this article, CIM-based DNN accelerators are designed and benchmarked under different chip area constraints. First, a scheduling strategy and dataflow for DNN inference is investigated when only part of the weights can be stored on-chip. Two weight reload schemes are evaluated: 1) reload partial weights and reuse input/output feature maps and 2) load a batch of input and reuse the partial weights on-chip across the batch. Then, system-level performance benchmark is performed for the inference of ResNet-18 on ImageNet data set. The design tradeoffs with different area constraints, dataflow, and device technologies [static random access memory (SRAM) versus ferroelectric field-effect transistor (FeFET)] are discussed.

DPAM: A Demand-Based Page-Level Address Mappings Algorithm in Flash Memory for Smart Industrial Edge Devices

Edge computing brings data storage closer to the location where it is needed. Therefore, the edge devices, especially smart industrial edge devices, require higher storage systems. NAND flash memory has the advantages of small size, high speed, and strong shock resistance, which is widely used in various storage systems, providing a good choice for edge devices. NAND flash has unique physical characteristics, such as “out-of-place updates” and “prewrite erasure,” therefore, the traditional address mapping methods require improvement. This article presents a novel demand-based page-level address mapping algorithm called DPAM. The goal of DPAM is to provide efficient address translation by using a smaller address mapping table. Due to the high service cost of block-level address mapping and hybrid address mapping, a page-level address mapping scheme is proposed. The algorithm is implemented and tested on the flash simulation platform FlashSim. The results indicate that our algorithm provides improvements of 7.11% for the hit ratio and 7% for the number of block erasures compared with other approaches.

Managing resources continuity from the edge to the cloud: Architecture and performance

The wide spread deployment of smart edge devices and applications that require real-time data processing, have with no doubt created the need to extend the reach of cloud computing to the edge, recently also referred to as Fog or Edge Computing. Fog computing implements the idea of extending the cloud where the ”things” are, or in other words, improving application performance and resource efficiency by removing the need to processing all the information in the cloud, thus also reducing bandwidth consumption in the network. Fog computing is designed to complement cloud computing, paving the way for a novel, enriched architecture that can benefit from and include both edge(fog) and cloud resources. From a resources perspective, this combined scenario requires resource continuity when executing a service, whereby the assumption is that the selection of resources for service execution remains independent of their physical location. This new resources model, i.e., resource continuity, has gained recently significant attention, as it carries potential to seamlessly providing a computing infrastructure from the edge to the cloud, with an improved performance and resource efficiency. In this paper, we study the main architectural features of the managed resource continuity, proposing the foundation of a coordinated management plane responsible for resource continuity provisioning. We study an illustrative example on the performance benefits in relationship to the size of databases with regard to the proposed architectural model.

Virtual Network Embedding for Collaborative Edge Computing in Optical-Wireless Networks

As an open integrated environment deployed with wired and wireless infrastructures, the smart city heavily relies on the wireless-optical broadband access network. Smart home data are usually sent to neighbor optical network units (ONUs) through front-end wireless mesh networks (WMNs) and finally reach the optical line terminal (OLT) for decision making via the passive optical network (PON) backhaul. To reduce backhaul bandwidth saturated by this conventional approach, smart edge devices (EDs) should be deployed at sensors and ONUs so that collaborative edge computing can be performed in front-end WMNs. Moreover, the cooperation of EDs at different ONUs is also promising for computing tasks that cannot be handled within front-end WMNs due to the local bottleneck, leading to collaborative edge computing in the PON backhaul. In this paper, network virtualization is utilized to support the coordination of computing and network resources. We also describe the relationship between virtual networks and requirements of computing tasks for substrate resources. First, a graph-cutting algorithm is employed to embed as many virtual networks as possible onto the common network infrastructure in front-end WMNs, aiming at minimizing the total transmitting power. Next, we transform impossibly embedded virtual networks into new ones that must be processed through the PON backhaul where the wavelength consumption will be optimized. Simulations results demonstrate that 1) the total transmitting power assigned for nodes is effectively reduced using the graph-cutting algorithm if all computing tasks can be solved by front-end WMNs; 2) otherwise, our method accepts more virtual networks with the improvement ratio of 77%, through the PON backhaul. In addition, there is a good match between the algorithm result and the optimal number of consumed wavelengths per optical fiber cable.

Supporting distributed feed-following apps over edge devices

In feed-following applications such as Twitter and Facebook, users (consumers) follow a large number of other users (producers) to get personalized feeds, generated by blending producers- feeds. With the proliferation of Cloud-connected smart edge devices such as smartphones, producers and consumers of many feed-following applications reside on edge devices and the Cloud. An important design goal of such applications is to minimize communication (and energy) overhead of edge devices. In this paper, we abstract distributed feed-following applications as a view maintenance problem, with the goal of optimally placing the views on edge devices and in the Cloud to minimize communication overhead between edge devices and the Cloud. The view placement problem for general network topology is NP Hard; however, we show that for the special case of Cloud-edge topology, locally optimal solutions yield a globally optimal view placement solution. Based on this powerful result, we propose view placement algorithms that are highly efficient, yet provably minimize global network cost. Compared to existing works on feed-following applications, our algorithms are more general--they support views with selection, projection, correlation (join) and arbitrary black-box operators, and can even refer to other views. We have implemented our algorithms within a distributed feed-following architecture over real smartphones and the Cloud. Experiments over real datasets indicate that our algorithms are highly scalable and orders-of-magnitude more efficient than existing strategies for optimal placement. Further, our results show that optimal placements generated by our algorithms are often several factors better than simpler schemes.